HelloFresh Adopts Tecton to Improve Customer Experience with Machine Learning

Machine Learning

HelloFresh, a fresh produce and meal kit delivery company, is enthusiastic about using machine learning to improve the customer experience. We had built quite a few machine learning and prediction systems in-house, but this homegrown approach was reaching its limits.

HelloFresh explored new options and recently chose Tecton’s feature platform for real-time machine learning.

“Prior to Tecton, our functions were spawned individually in individual Spark pipelines. We lacked the ability to provide functionality for reasoning,” said Benjamin Bertincourt, senior manager of ML engineering at HelloFresh SE.

HelloFresh aimed to make machine learning more standardized and available at scale. The feature store was a key component of HelloFresh’s planned approach.

in a machine learning system feature A type of variable used as input for predictive models such as fraud detection and recommendation engines. Characteristics include how much the customer has purchased in the last 30 days, the current price of the item, and whether the item is in stock.

Without a feature store, the latest information would have to be retrieved from the raw data system, processed, and then used, which would slow everything down. Functionality can be abstracted from raw data, providing a more consistent approach across different systems and making it easier to share high-quality functionality across teams.

“We are in the process of switching our primary production models to be sourced from the Tecton functional store. It allows us to reuse the functionality that we have already created.” Bertincourt. “One team is currently using a generalized embedding model that uses features provided by Tecton, with the aim of making the embeddings accessible and usable in machine learning models. , providing a foundation for deploying predictive and personalization models in the future.”

The feature store can also provide important governance information such as the version of the feature used to make a particular prediction. This is essential for model debugging and compliance with some industry regulations. Good governance tools are especially important for online systems, because good predictions require current, high-quality data, rather than static collections of old, stale data of dubious provenance.

“We are focused on collecting data to improve how our customers interact with our products and to use customer data judiciously,” says Bertincourt. “Our data collection is entirely focused on helping us model the customer experience, rather than collecting large amounts of data just because we can.”

As companies’ data practices come under increasing scrutiny, careful use of data should be a priority for every data science team. As teams build highly automated inference systems, they will need more tools like Tecton. Customers want confidence that companies are only using ethically collected data for their own benefit. Feature stores like Tecton are one way businesses can demonstrate to customers and regulators that machine learning is being used responsibly, while improving the customer experience.

Follow me please twitter or LinkedIn. check out my website.

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *